Beta 3

Released Beta 3 version

To get started:

1. Select app based on need

There are 4 Linux Binary containing:
aarch64 for arm computing
normal-x86_64 for integrated graphic
nvidia-x86_64 for Nvidia GPU acceleration
amd-x86_64 for AMD radeon GPU acceleration

There are 3 Window Binary containing:
normal-x86_64 for integrated graphic
nvidia-x86_64 for Nvidia GPU acceleration
amd-x86_64 for AMD radeon GPU acceleration

There is only one Mac binary which has Metal GPU acceleration support

2. Downloading the favorable LLM AI Model
   Press F2 to open Registry Menu and type model name to search and then use up/down arrow key to navigate. Press enter to Download.
   Caution: Instruct model may ignore the system tool instruction

3. Selecting Model
   After Downloading model, select the downloaded model by pressing F3 to open Modal menu

4. Configuration (Optional)

• Power Mode: to select the how much power want to spend on this AI model
• MTP: multi tokens prediction is the special feature that create extraction prediction of next tokens with the cost of RAM usage and precision.
• Auto Collapse: If enable then every Agent response will collapse previous label
• Target FPS: for feeling smooth animation then you can increase the FPS
• Stall Time: In some worse case like AI stuck on prefill then Stall watchdog help to prevent this issue
• Repeat Detector: In worst case, AI may hallucinate and repeat tokens continuously. So system detect the consecutive repeated text and notify the AI about repeatation.
• Context Window: In worst case, Application may crash cause of KV cache overneed. Some OS may kill/end that process for system safety. So decreasing Context Window may help to prevent of this causes.
• Permission: To give AI allowance to do action and directory access. by default always allow
• Web Search: To configure which Web provider you let AI to search online. By default DDG
• HF Token: In some case, you tried to search but got empty result that may cause of block from Hugging Face for too many request. so create the HF Token and paste it on to avoid this empty error.

Released Beta B2

Full Changelog: v0.0.1b...v0.0.2b

Beta B1

After entering Hercules Agent,

Hold ESC to exit (to prevent accidentally click)

To download local model from settings, press F2

show/hide log activity, press F3

F1 for help

To get Started: press F2 and choose any AI model feasible for your hardware.

There are two main repos:

• Ollama ( Require to install ollama )
• HuggingFace ( used llama.cpp which is embed on it and best choice for performance )